Geophysical Research Letters

Effects of fire-precipitation timing and regime on post-fire sediment delivery in Pacific Northwest forests



[1] Wildfires affect the coupled dynamics of vegetation, runoff response, and sediment production, as well as the sequencing of post-fire precipitation and snowmelt in forested watersheds. We examined these interactions by applying a spatially distributed hydrologic model to multiple-year periods before and after a major fire that occurred in 1970 in the Entiat River basin, Washington. The effects of precipitation sequencing on post-fire sediment delivery were examined by simulating the 1970 fire as if it had occurred at other times in a 50-year period. Simulated sediment delivery varied by a factor of two depending on fire timing. We also compared the effects of fire suppression and found that simulated sediment production was about 20% higher for natural compared with current conditions.

1. Introduction

[2] Wildfires can change sediment delivery to stream channels by altering vegetation and soil [Benda et al., 2003; Cannon et al., 2001]. Vegetation mortality lowers slope stability because it decreases root cohesion and transpiration, thereby increasing soil moisture and pore pressure. Surface erosion can increase post-fire due to higher precipitation throughfall and increased overland flow associated with soil hydrophobicity [DeBano, 2000]. We expand on earlier studies of the interaction of fire, sediment generation, and storm characteristics [e.g., Benda and Dunne, 1997; Gabet and Dunne, 2003; Istanbulluoglu et al., 2004] to represent the time-varying dynamics of post-fire sediment generation using an explicit spatially distributed model.

2. Methods

2.1. Model and Experiment Description

[3] We used the Distributed Hydrology-Soil-Vegetation Model (DHSVM) [Wigmosta et al., 1994] and its sediment model [Doten et al., 2006] to study the effects of fire on sediment generation in the eastern Cascade Mountains, Washington. DHSVM has been widely used to predict the hydrologic effects of land use change on forested catchments in the Pacific Northwest [e.g., Alila and Beckers, 2001]. The sediment model predicts mass wasting, surface erosion from hillslopes and roads, and sediment transport via channel routing. The distributed nature of DHSVM is ideal for addressing fire impacts on aquatic habitat size, quality, and connectivity. We altered DHSVM to model the effects of post-fire vegetation and soil disturbances.

[4] As in work by Agee [1993], we defined three fire regime categories. Low severity fire regimes are typical of the lower elevation eastern Cascade forests; fires occur fairly frequently but mainly affect understory vegetation. Forests in high severity fire regimes burn rarely, but fires kill both understory and overstory. Mixed severity fire regimes occur at intermediate elevations and tend to have patches of high and low intensity during individual fires; fire frequencies fall between low and high severity fire regimes. For each severity class, we define a mean fire interarrival time.

[5] We modeled fire effects on catchment hydrology (and its corollary, sediment production [see Doten et al., 2006]), by adjusting three parameters in DHSVM: leaf area index (LAI), root cohesion, and maximum infiltration rate. Root cohesion and LAI were updated based on estimates for coastal Douglas Fir by Sidle [1992] (see auxiliary material). We modeled the effects of hydrophobicity on grid cells where the pre-fire vegetation type was conifer forest [see DeBano, 2000] by lowering the maximum infiltration rate to 33% of its pre-fire value after a high severity fire, 66% after a medium severity fire, and 75% after a low severity fire. We then increase this rate linearly (results for exponential and parabolic increases were similar) for six years to its pre-fire value, following observations of hydrophobic persistence in the Cascades [Dyrness, 1976].

2.2. Model Implementation

[6] The Entiat River basin (EB) is located in north-central Washington on the eastern slopes of the Cascade Mountains (Figure 1). This 527 km2 drainage area upstream of USGS gage 12452800 is composed of metamorphic schist and gneiss, granodiorite, and quartz diorite, covered by volcanic ash and pumice [U.S. Department of Agriculture (USDA), 1979]. Annual average precipitation, 75% of which falls between October and March, ranges from about 2.3 m in the headwaters to less than 0.5 m in the lower reaches.

Figure 1.

Location of EB and DEM of the basin above the USGS gage at Ardenvoir. The basins of EEW are outlined.

[7] Within EB, three small catchments (Fox, Burns, McCree) comprise the Entiat Experimental Watersheds (EEW), each draining about 5 km2. The U.S. Forest Service measured EEW streamflow and sediment delivery from 1961–1977 [Helvey, 1980]. On 24 August 1970, high severity wildfires burned 22% of EB, including the entire EEW [USDA, 1979].

[8] We applied DHSVM at a 3-hour time step (appropriate for snowmelt-dominated systems that are controlled by accumulated winter precipitation) over EEW at 30-m spatial resolution for the period 1964–1982 and over EB at 90–m spatial resolution from 1930–1990. As in the work by Doten et al. [2006], we used a 10-m DEM for mass-wasting calculations. Model climate forcings were taken from Hamlet and Lettenmaier [2005], which includes topographic adjustments to precipitation and temperature, adapted to local observations from the Pope Ridge NRCS SNOTEL station. We divided daily precipitation evenly into 3-hour time steps. Pre-fire vegetation data were taken from the potential vegetation types (based on environmental conditions, such as topography, precipitation, and temperature) of the Washington GAP project [Cassidy, 1997]. Soils data were modified from Doten and Lettenmaier [2004] (Table S1).

[9] Soil lateral hydraulic conductivity and the parameter that represents changes in hydraulic conductivity with depth in DHSVM were adjusted to match observed EEW streamflow for the pre-burn period 1964-1970. Soil cohesion values were adjusted to estimate observed pre-fire sediment delivery. Soil depth was taken from soil maps.

[10] To model precipitation-fire-erosion interactions, we used a shuffled-deck approach, consisting of 30 simulations in which we rotated the burn year through the period 1952–1981. Fires were assumed to start on 24 August of the first year of each 20-year simulation.

[11] The 1970 fire in the EEW uniformly burned at a high severity; however, over a larger area, fire regimes vary in severity with vegetation type and location [Agee, 1993]. In anomalously warm, dry periods, fire regimes shift to higher severity [e.g., Gedalof et al., 2005; Meyer and Pierce, 2003]. Similarly, decades of fire suppression have shifted low and mixed fire regimes to higher severity [Agee, 1993].

[12] To explore fire regime shifts, we generated plausible fire scenarios over EB with a stochastic fire model (see auxiliary material). Following Agee [1993] and Benda and Dunne [1997], we modeled fire frequency as a Poisson process. We treated fire size for each model fire probabilistically, using the 433-year fire history of the Teanaway basin in the eastern Cascades [Wright and Agee, 2004]. Severity of the fire placed upon each pixel was based on the fire regime and vegetation type of that pixel. We generated three fire sequences (“scenarios”) for current and natural conditions (CC and NC, respectively), described below, for the period 1930–1980 (see Figure 4). We then ran DHSVM, changing vegetation and other fire-related model parameters for each of six scenarios, until 1990, at least 10 years after the last fire.

[13] Using the stochastic fire generator in EB, we modeled scenarios reflecting the lower severity, higher frequency and more spatially heterogeneous NC of the 18th to early 20th centuries. Each potential vegetation type [Cassidy, 1997] was assigned to the low, mixed, or high severity regime following Agee [1993]. The fire regime map was adjusted based on elevation, topography and nearby fire regimes, as well as a personal communication with J. K. Agee (2005).

[14] We modeled CC scenarios as less frequent and more severe than those of NC, reflecting decades of fire suppression. The CC fire severity fractions were based on observed fire intensities during the 1994 Tyee fire that burned much of EB (J. K. Agee, unpublished report, 1995).

[15] Interarrival times for NC were calculated from reconstructed historical fire timing and extent histories from Schellhaas et al. [2001] for high severity pixels, and from Everett et al. [2000, 2003] for low severity pixels. Mixed severity interarrival times were taken from Agee [1993, 2005].

3. Results

3.1. Model Testing

[16] DHSVM captured general EEW post-fire trends in runoff, including post-fire dampening of the diurnal flow cycle and dramatic soil moisture increases (Figures S2 and S3) [Helvey, 1980]. The modeled post-fire hydrology matched observed runoff timing well but overestimated peak flows (Figure 2). Debris flows in 1972 compromised post-fire streamflow records in two (Fox and McCree) EEW catchments. Burns data quality was considered good [Woodsmith et al., 2004]. Given the issues with post-fire streamflow records, and the fact that most of the simulated erosion is driven by mass wasting (Figure S4), the overestimation of flow may be inconsequential to the discussion that follows.

Figure 2.

(a) (left) Daily streamflow and (right) mean monthly streamflow before and after 24 August 1970 fire for EEW. “Modeled, fire” includes vegetation and soil disturbances. Observations were unavailable in 1973 for Fox and 1972 for McCree Creeks. Mean monthly streamflow color scheme matches that of daily streamflow. (b) Modeled sediment delivery with and without the 24 August 1970 fire, for low, medium and high severity fires, at the outlets of EEW. Observations were unavailable for 1973–1974. Total erosion equals surface erosion plus mass wasting delivered to the channel network. Modeled sediment is about an order of magnitude higher in post-fire years for high severity than for no fire.

[17] Due to the high infiltration capacity of soils in EEW, predicted sediment generation was insensitive to hydrophobic effects and instead was dominated by changes in vegetation – in particular, overstory LAI. Overstory removal decreases snow interception and ablation, yielding more spring snow accumulation [see Storck et al., 2002]. The overstory also attenuates solar radiation; thus, its removal enhances peak melt rates. Although complete overstory mortality was observed, initially setting LAI to zero may be unrealistic since vegetative debris intercepts snow and attenuates solar radiation to some extent.

[18] Figure 2 shows modeled sediment generation for uniformly low, medium and high severity fires for each of the 30 years. Helvey [1980] reports that the 1970 fire burned at uniformly high severity across EEW, and total observed and predicted sediment match best for a case intermediate between medium and high severity (Figure 2).

[19] After the fire, sediment from the weir ponds was measured annually. The escape of sediment from the weir ponds makes observations conservative [Helvey, 1980]. Large debris torrents on Fox and McCree Creeks in 1972 were not trapped, but Helvey [1980] estimated the sediment volume left in alluvial fans. Modeled sediment delivery excludes channel scour and deposits debris flow sediment in channels at shallow slopes [Doten et al., 2006]. Since observed values are conservative, the model results seem plausible.

3.2. Precipitation Interactions

[20] Figure 3a shows EEW modeled annual sediment delivery for the shuffled-deck experiments. The years 1968, 1972, and 1974 produced large sediment volumes for most fire dates. All three had ample snowpack to saturate soils. DHSVM predicted the largest 20-year sediment delivery for fires occurring in the 1960s, with peak sediment delivery after a presumed 1964 fire (Figure 3b). The period of reduced root cohesion after a 1964 fire coincides with three high snowpack years, with the minimum root cohesion in 1972, the year of the highest snowpack of record. Sediment generated during the first 20 post-fire years in EEW varied by more than a factor of two depending on the timing of fire relative to climatology.

Figure 3.

(a) Annual sediment delivery in EEW for 20 years after the fire for experiments in which the 1970 fire was assumed to occur in each year from 1952–1981. Time series are color-coded by date of fire, e.g., light green corresponds to fire occurring in 1952; bright red in 1962, light blue in 1972, and so on. (b) Total delivered sediment summed over 20 post-fire years for each time series in Figure 3a, plotted at the year of fire initiation. Color code matches Figure 3a. Black portion of bar shows surface erosion, colored mass wasting. Sediment delivered in 1972 is plotted as a function of the year of disturbance.

3.3. Effect of Fire Regime Changes

[21] Figure 4 shows that sediment tended to come in large pulses for CC, similar to observations by Meyer and Pierce [2003]. Because overland flow is driven by saturation excess runoff in this region, the year with the highest basin saturation, 1972, produced the largest surface erosion events for all scenarios (Figure S5). CC scenarios 4 (19,500 ha in 1971) and 5 (11,100 ha in 1967) had large burned areas preceding the 1972 storms; yet, despite lower root cohesion in 1972, scenario 5 generated less sediment because more sediment was removed earlier in the sequence (Figure 4b).

Figure 4.

(a) (left to right) Burn severity area maps, number of slope failures maps, and number of grid cells experiencing a given number of slope failures and (b) fire history sequences with annual delivered sediment (black) for scenarios applied to EB. Scenarios labeled 1–3 are under NC and 4–6 are under CC.

[22] Mass wasting dominated modeled sediment generation for all scenarios (Figure S5). Surface erosion may have been somewhat underestimated because the 3-hour time step masks peak precipitation during infrequent but intense localized summer thunderstorms. In total, the NC scenarios generated about 20% more sediment than the CC scenarios. Although simulated sediment generation was higher for individual fires in CC, their lower frequency resulted in less sediment overall. Although based on plausible combinations of fire statistics and physical implications of fires of varying severity, these results could well differ under other combinations of climate, vegetation, catchment physical characteristics.

4. Conclusions

[23] Application of our spatially distributed hydrology and sediment model resulted in plausible simulations of post-fire sediment generation in EEW. In our simulations, loss of vegetation strongly impacted runoff and sediment generation in this snowmelt-dominated system. Post-fire increases in sediment generation were primarily related to slope failures, and were strongly controlled by the interactions of post-fire loss of root cohesion with spring snowmelt. Spring snowmelt changes in turn were closely linked to reduction in overstory LAI, which reduced winter ablation and hence increased spring snow accumulation, and increased peak spring melt rates due to reduction in canopy attenuation of solar radiation. High sediment generation was critically dependent on interactions of high snow accumulation years during the period when post-fire root cohesion was at its lowest.

[24] In larger area EB simulations, NC yielded about 20% more sediment than CC, because increased frequency of fire events in NC more than compensated for increased fire severity in CC. This result is due in part to specification of fire severity spatially in the experiments; high elevation areas, which have steep slopes and contribute much of the basin sediment yield, burn at high severity in both CC and NC, although at higher frequency in NC. As a result, high elevation areas produced sediment at a higher rate for longer periods in NC. Clearly, these results reflect a balance between fire statistics (interarrival time and mean area burned) and the effects on vegetation root cohesion and LAI of fires of varying severity. This balance could well differ under other combinations of climate, vegetation, and catchment physical characteristics. Future work should further explore this balance. Moreover, the effects of forest roads on sediment – not considered in this study – may also interact with the effects of fire suppression.

[25] Finally, sediment generation in CC tends to occur in infrequent, large bursts, whereas more frequent, low level sediment peaks are generated under NC. In-stream disturbances in CC are more likely to be short relative to the lifespan of the longest living individuals of a fish population, increasing the rate of recovery; however, the volume of sediment input may impede movement into and out of the impacted region, slowing recovery [Detenbeck et al., 1992]. These implications likewise are worthy of further investigation.


[26] Funding for this research was provided by the USFS Pacific Northwest Research Station and Wenatchee Forestry Sciences Laboratory under Research Joint Venture agreement 00-RJVA-11261927-522.